Design And Implementation Of A RAG-Based Personalized Nutrition Advisory System With Multi-Parameter Health Risk Prediction Using Machine Learning

Abstract

Dietary management and preventive health monitoring are critical components of modern healthcare, yet existing systems lack the ability to provide personalized, evidence-grounded nutritional guidance based on individual health parameters. This paper presents the design and implementation of a Retrieval-Augmented Generation (RAG) based personalized nutrition advisory system integrated with multi-parameter health risk prediction. The proposed system employs a curated nutrition knowledge base derived from WHO and ICMR dietary guidelines, combined with a large language model inference pipeline using the Groq API (llama-3.3-70b-versatile), to generate contextually accurate dietary recommendations. The system was implemented and tested with real users, generating complete personalized meal plans based on actual recorded vitals (HR=78 BPM, BP=100/65 mmHg, BMI=16.9 Underweight). The health risk prediction module computes weighted risk scores for diabetes, hypertension, heart disease, and obesity based on 14-day dietary pattern analysis from MongoDB. Additional components include a food photograph recognition module (successfully identifying Misal Pav at 550 kcal with full macronutrient breakdown), a BMI calculator using Mifflin-St Jeor formula (TDEE=1895 kcal, Goal=2195 kcal), and a gamification engine with streak-based behavioral incentives. Built on the MERN stack, the system demonstrates practical feasibility for preventive nutrition management accessible through any standard web browser.